Mohamed Anis Loghmari

The contribution of the sources separation method in the decomposition of mixed pixels

In this paper, we propose to prove the importance of the application of blind sources separation methods on remote sensing data. Indeed, satellite images are represented by radiometric values where each one is considered as a mixture of different sources. The primary goal of our research is to hand back the different sources covering the scanned zone. The main constraint to restore these sources is to take our observation images as a mixture of physically independent components. In our work, the independence between the different sources is obtained through two statistical methods. The first method is based on the reduction of the spatial source correlations, and the second one is based on the joint maximization of the fourth-order cumulants. On the opposite of the original multispectral images that are represented according to correlated axes, the source images extracted from the proposed algorithms are represented according to mutually independent axes that allow each source to represent specifically a certain type of land cover. This increases the reliability of the analysis and the interpretation of the scanned zone. The source images obtained from the application of the sources separation method give a more effective representation of the information contained on the observation images. The performance of these source images is investigated through an application for the decomposition of mixed pixels. The originality of our application comes from the determination of the mixing matrix modeling the spectral endmembers based on source filters. These filters model the sensibility of each source channel according to the different spectral bands, which give an interesting information about the spectral theme represented by the corresponding source image. This application shows that the proportions of the different land cover types existing into the pixel are better estimated through the source images than through the original multispectral images. This method could offer an interesting solution to mixed-pixel classification.

A Spectral and Spatial Source Separation of Multispectral Images

This paper deals with the problem of blind source separation of remote sensing data based on a Bayesian estimation framework. We consider the case of multispectral images in which we have observed images of the same zone through different spectral bands. The land cover types existing in the scanned zone constitute the sources to separate. Associating each source to a specific significant theme remains the real challenge in the source-separation method applied to satellite images. In fact, multispectral images consist of multiple channels, each channel containing data acquired from different bands within the frequency spectrum. Since most objects emit or reflect energy over a large spectral bandwidth, there usually exists a significant correlation between channels. This constitutes the first difficulty for sources identification. The second difficulty lies in the heterogeneity of most of the geological and vegetative ground surfaces. In this case, the geometrical projection of a single detector element at the Earths surface, which is sometimes called the instantaneous field of view, is formed from a mixture of spectral signatures. In such circumstances, the needed information is either not available or not reliable. In this paper, the goal is to establish a new approach based on a two-level source separation (TLSS), which consists of a spectral separation along the different used bands and a spatial separation along neighboring pixels of each image band. The spectral separation has been used prior to the Bayesian approach, and it is based on a second-order statistics approach that exploits the correlation through different spectral bands of the multispectral sensor. The given images are represented according to independent axes that provide more effective representation of the information within the observation images. The spectral separation consists of identifying the sources without resorting to any a priori information, hence the term blind. The obtained sources represent the starting point for the Bayesian approach, which is known for its weakness in front of initial conditions. To identify a significant theme for each source, we have to spatially separate each image based on a Bayesian source-separation framework. The proposed approach has the added advantages of the blind source method as well as the Bayesian method. It should give segmented images related to each theme covering the scanned zone, which are the TLSS results of the observation images

A New Sparse Source Separation-Based Classification Approach

In many geoscience applications, we have to convert remotely sensed images to ground cover maps. Numerous approaches to extract ground cover information have been developed. Recently, blind source separation (BSS) of remote-sensing data has received significant attention due to its suitability to recover sources when no information is available about the scanned zone, hence the term blind. In the remote-sensing context, associating each source to a significant land cover theme is difficult and constitutes the real challenge of this paper. Many authors have pointed out that BSS is overwhelmingly a question of contrast and diversity. This reasoning motivates this work which takes advantage of both decorrelation and sparsity to propose a two-level novel approach to separate our different land covers called sources. The first separation stage is based on second-order statistics or decorrelation. It gives a suitable representation of the remote-sensing images. However, decorrelation is a natural way of differentiating statistically between sources but is unable to identify and extract finer features with physical meaning. The aim of the second separation stage is to overcome this problem by an increasingly popular and powerful assumption which is the sparse representation. The last leads to good separation because most of the energy in the defined basis, at any time instant, belongs to a single source. This allows the extraction of physical features and the capture of image essential structures. The innovative aspect of this study concerns the development of a new image classification approach that integrates the BSS at the feature extraction level to provide the most relevant sources from remotely sensed images. It can be viewed as an unsupervised classification method. The second-order separation process is used as a preprocessing step to remove the interband correlation which sometimes brings ill effect to image classification. However, the second-order process is unable to uncover the underlying sources. The basic idea behind our approach is that heterogeneous multichannel data provide sparse spectral signatures in addition to sparse spatial morphologies in specified dictionaries. Hence, sparse modeling can be used to disentangle the land covers from observed mixtures. From the sparse representation, the data space is transformed into a feature space composed of mutually exclusive classes. Finally, we will merge these classes at the decision level in order to enhance the semantic capability and the reliability of land cover classification. The effectiveness of the proposed approach was demonstrated by operating two experiments to study respectively the source separation and the image classification capability of the developed approach. The different results on remote-sensing images illustrate the good performance of the new sparse approach and its robustness to noise. These experiments show that the sparse representation enhances the separation quality and allows extracting more easily the essential structures of the scanned zone. The proposed approach offers an interesting solution to the classification process with limited knowledge of ground truth.

Mixed pixel decomposition of satellite images based on source separation method

In this paper we propose to prove the importance of the application of blind source separation methods on remote sensing data. Satellite images are represented by radiometric value that can be considered as a mixture of independent sources. To restore the independent sources we use the statistical method of Joint Approximate Diagonalization of Eigen-matrix (JADE). The proposed algorithm generates source images where each one gives a maximum of information specific to a certain type of land cover. These source images do not provide one scalar value per pixel, but rather a vector which components will agree with the radiometric value of the different land cover types present in the pixel.

Nonlinear Source Separation Based on Multi-Layer Perceptron: Application on Remote Sensing Analysis

Source separation is relatively a new area of data analysis. The most widely used separation approachs are linear. However, in many realistic cases the process which generates the observations is nonlinear and no information is available about the mixture. In this case, it can be expected to capture the structure of the data better if the data points lie in a nonlinear manifold instead of a linear subspace. In this paper, we try to find a model which allows a compact description of the observations in the hope of discovering some of the underlying causes or sources of the observations. Then, we will process a dimension reduction to classify the obtained sources and evaluate the performances of the proposed method.

An efficient wavelet dictionary for texture separation

In this paper, our goal is to highlight the importance of the source separation method on remote sensing data analysis when dealing with urban areas characterized by spatial concept like texture. Source separation has become an attractive tool used to compensate physical information deficiency by statistical assumptions. The methods key comes from the fact that the blind signal separation can be achieved by restoring statistical independence. In this work, we try to design a statistical generative model, based on a wavelet dictionary, composed of atoms which are automatically selected to maximize the sparseness of the underlying texture type. This application is of utmost importance in the classification process and should minimize the misclassification risk of urban areas.

Parametric versus discriminative learning approach for multispectral image recognition

Land cover classification for remote sensing data have motivated several researches such us source separation, feature extraction and classification method. In this work, we aim to provide enhanced pattern recognition method based on source separation. Then, new data presentation is processed by feature extraction and adaptative classification. The non linearity for the mixing phenomenon is approximted by neural network. Adaptative classification approch is based on a parametric feature characterization. We will compare this approch to a discriminative classifier such as Support vector machine. Experimentations are based on SPOT4 observations. They demonstrates that the new presentation and the parametric model allows more efficient pattern identification. The results prove the potential of the method for urban areas.

A new classification approach based on source separation and feature extraction

Pattern recognition for multispectral data aims to identify land cover thematics for environmental monitoring and disaster risk reduction. Multispectral images contain data acquired from different channels within the frequency spectrum. They represent a mixture of latent signals. This paper represents a pattern recognition contribution for remote sensing. We propose a new classification framework based on nonlinear source separation and linear feature fusion. The first stage performs a nonlinear separation model based on multilayer neuron network. The underlying sources are Gaussians and a misfit function between the approximated source distributions and their priors will be minimized iteratively. The second stage performs feature extraction and fusion. The linear feature model considers that feature descriptors allow cooperative description for land pattern recognition. Classification tasks are performed by Support Vector Machine. Experimentation results demonstrate that the proposed classification method enhances the recognition accuracy and provides a powerful tool for land identification.

Contribution of the Methodologies Development for the Analysis of Remote Sensing Data

In this paper, we consider the problem of Blind source separation (BSS) method by taking advantage of the sparse modeling of the hyperspectral images. These images are produced by sensors which provide hundreds of narrow and adjacent spectral bands. The idea behind transform domains is to apply some transformations to illustrate the dataset with a minimum of components and a maximum of essential information. To take advantages from the new representation of hyperspectral data, a novel classification approach based on using Binary Partition Trees (BPT). The BPT is obtained by iteratively merging regions and provided a combined and hierarchical representation of the image in a tree structure of regions.